Does Pgbouncer answer to kill command? - postgresql

Question

I have configured Pgbouncer as a sidecar pattern in one of my pods in Azure Kubernetes based on Azure Oss Db Tools Pgbouncer Sidecar documentation. It has the following container lifecycle hook:
lifecycle:
preStop:
exec:
command: ["/bin/sh", "-c", "killall -INT pgbouncer && sleep 120"]
I believe the intended purpose of this command is to wait 120 seconds until any running query finishes.
To understand what it does, I opened two interactive shells inside the Pgbouncer container. In the first shell, I executed the killall command and in the second shell, I executed ps command multiple times.
First shell:
/ $ ps
PID USER TIME COMMAND
1 postgres 0:00 /usr/bin/pgbouncer /etc/pgbouncer/pgbouncer.ini
6 postgres 0:00 /bin/sh
30 postgres 0:00 ps
/ $
/ $
/ $ killall -INT pgbouncer && sleep 120
Second shell:
/ $ ps
PID USER TIME COMMAND
1 postgres 0:00 /usr/bin/pgbouncer /etc/pgbouncer/pgbouncer.ini
6 postgres 0:00 /bin/sh
33 postgres 0:00 /bin/sh
40 postgres 0:00 sleep 120
41 postgres 0:00 ps
/ $
/ $
/ $ ps
PID USER TIME COMMAND
1 postgres 0:00 /usr/bin/pgbouncer /etc/pgbouncer/pgbouncer.ini
6 postgres 0:00 /bin/sh
33 postgres 0:00 /bin/sh
42 postgres 0:00 ps
After 120 seconds, Pgbouncer main process is still running (See the output from the second shell). I thought this should have terminated both my terminal sessions since it was supposed to kill Pgbouncer process (PID = 1) and stop the container.
If I try to kill using the below command:
/ $ kill 1
/ $ command terminated with exit code 137
I see that both my terminal sessions are immediately terminated and the container is stopped.
I want to understand whether we really need this lifecycle hook since it is not properly working? Or did I make any mistake trying to understand what it does?
Thanks for the help! 🙏

Answer 1

There is a difference here. The killall -INT sends INT signal while kill sends TERM signal if no signal is specified. You can try again with kill -INT 1 to see whether it gets the same behavior. I think the pgbouncer process is also catching INT.
Here is reference to the site:
int cf_shutdown; /* 1 - wait for queries to finish, 2 - shutdown immediately */
...
static void handle_sigterm(evutil_socket_t sock, short flags, void *arg)
{
log_info("got SIGTERM, fast exit");
/* pidfile cleanup happens via atexit() */
exit(1);
}
static void handle_sigint(evutil_socket_t sock, short flags, void *arg)
{
log_info("got SIGINT, shutting down");
sd_notify(0, "STOPPING=1");
if (cf_reboot)
die("takeover was in progress, going down immediately");
if (cf_pause_mode == P_SUSPEND)
die("suspend was in progress, going down immediately");
cf_pause_mode = P_PAUSE;
cf_shutdown = 1;
}
Pgbouncer stops processing any more queries when the INT signal is given while the TERM signal will terminate the process immediately.